When the repository writes the prompt: hijacking AI coding agents
How instructions planted in READMEs, issue templates and test fixtures steer AI coding agents into leaking secrets, and the guardrails that held.
Vulnerability research & advisories
Original research and coordinated disclosure across cloud, AI, OT and enterprise software. Every advisory, every CVE, in one place.
Advisories
Every vulnerability we disclose gets a TRA: what is affected, how severe it is, and where the fix stands. Filter by severity or search a CVE, vendor or product.
No advisories match these filters.
Try a CVE identifier, a vendor name or a broader severity.
Showing 8 of 8 latest advisories
Research
How instructions planted in READMEs, issue templates and test fixtures steer AI coding agents into leaking secrets, and the guardrails that held.
Coverage
Our researchers specialise by attack surface, from cloud control planes to the controllers on a factory floor.
Coordinated disclosure
Every vendor gets the same window to ship a fix. When it closes, defenders get the details, patched or not.
We send the vendor full technical details and a proof of concept over a secure channel.
We expect an acknowledgement within a week, and help reproduce if needed.
Midway, we confirm a fix date. Credible progress can earn a short extension.
The TRA goes live with details, CVEs and the vendor’s remediation guidance.